Add mechanism for master and worker nodes to infer the JVM memory settings based on cgroups

[tmckayus]

Closing this pull request, which was based on https://github.com/radanalyticsio/openshift-spark/pull/71

This change is not needed in newer versions of Java (see notes on 71)